name: create-skills description: This skill should be used when the user asks to "create a skill", "build a skill", "write a skill", "improve skill structure", "understand skill creation", or mentions SKILL.md files, skill development, progressive disclosure, or bundled resources (scripts, references, assets).

Every skill created should follow pure XML structure, enforce progressive disclosure, and apply expert prompting techniques from create-prompt skill.

<quick_start>

Identify the reusable pattern: What procedural knowledge would help future tasks?
Create directory structure: mkdir -p ~/.claude/skills/{skill-name}/{references,scripts}
Write SKILL.md:
- YAML frontmatter (name, description in third person)
- Pure XML body structure (no markdown headings)
- Required tags: <objective>, <quick_start>, <success_criteria>
Add bundled resources: Scripts, references, or assets as needed
Test with real usage: Iterate based on observations
Create slash command wrapper: Optional wrapper for discoverability

<minimal_example>

---
name: process-pdfs
description: Extract text and tables from PDF files, fill forms, merge documents. Use when working with PDF files or when the user mentions PDFs, forms, or document extraction.
---

<objective>
Extract text and tables from PDF files, fill forms, and merge documents using Python libraries.
</objective>

<quick_start>
Extract text with pdfplumber:

```python
import pdfplumber
with pdfplumber.open("file.pdf") as pdf:
    text = pdf.pages[0].extract_text()

</quick_start>

<success_criteria>

Text extracted successfully from PDF
Output is clean and properly formatted
No encoding errors </success_criteria>

</minimal_example>
</quick_start>

<core_concepts>
<skills_anatomy>
Every skill consists of:

**SKILL.md (required)**:
- YAML frontmatter: name, description
- Pure XML body: no markdown headings (#, ##, ###)
- Required tags: objective, quick_start, success_criteria
- Conditional tags based on complexity

**Bundled Resources (optional)**:
- **scripts/**: Executable code for deterministic tasks
- **references/**: Documentation loaded as needed (progressive disclosure)
- **assets/**: Files used in output (templates, boilerplate)
</skills_anatomy>

<progressive_disclosure>
Skills use a three-level loading system:

1. **Metadata (name + description)** - Always in context (~100 words)
2. **SKILL.md body** - When skill triggers (<500 lines ideal)
3. **Bundled resources** - As needed by Claude (unlimited)

**Critical principle**: Keep SKILL.md lean (<500 lines). Move detailed content to references/.
</progressive_disclosure>

<when_to_create_skills>
**Create skills for**:
- Reusable patterns across multiple tasks
- Domain knowledge that doesn't change frequently
- Complex workflows with structured guidance
- Reference materials (schemas, APIs, libraries)
- Validation scripts and quality checks

**Use prompts for**: One-off tasks that won't be reused

**Use slash commands for**: Explicit user-triggered workflows with fresh context
</when_to_create_skills>
</core_concepts>

<skill_creation_process>
<step_1>
**Understanding Use Cases with Concrete Examples**

To create an effective skill, clearly understand how it will be used:

- What functionality should the skill support?
- Can you give examples of how this skill would be used?
- What would a user say that should trigger this skill?

Conclude when you have a clear sense of functionality the skill should support.
</step_1>

<step_2>
**Planning Reusable Skill Contents**

Analyze each example by:

1. Consider how to execute from scratch
2. Identify helpful scripts, references, and assets

**Example patterns**:

- **Scripts**: Rotating PDFs requires re-writing code → `scripts/rotate_pdf.py`
- **Assets**: Frontend apps need boilerplate → `assets/hello-world/` template
- **References**: BigQuery needs table schemas → `references/schema.md`

Create a list of reusable resources to include.
</step_2>

<step_3>
**Creating Skill Structure**

```bash
mkdir -p ~/.claude/skills/{skill-name}/{references,scripts}
touch ~/.claude/skills/{skill-name}/SKILL.md

Directory naming:

Follow verb-noun convention: create-*, manage-*, setup-*, generate-*
Lowercase letters, numbers, hyphens only
Must match name in YAML frontmatter exactly </step_3>

<step_4> Writing SKILL.md with Expert Prompting

Apply create-prompt principles:

Use the create-prompt skill to ensure your SKILL.md follows expert prompting techniques:

Clear, specific instructions
XML tags for structure
Success criteria defined
Examples when format matters
Context management for long-running tasks

YAML frontmatter requirements:

---
name: skill-name-here
description: This skill should be used when the user asks to "specific phrase 1", "specific phrase 2", or mentions "trigger words". Include what it does AND when to use it.
---

Critical rules:

Third person: "This skill should be used when..." (never "Use this skill...")
Specific triggers: Include exact phrases users would say
Maximum 1024 characters
No XML tags in description

Good description examples:

description: This skill should be used when the user asks to "create a hook", "add a PreToolUse hook", "validate tool use", or mentions hook events (PreToolUse, PostToolUse, Stop).

Bad description examples:

description: Use this skill when working with hooks.  # Wrong person, vague
description: Provides hook guidance.  # No trigger phrases

Pure XML body structure:

Remove ALL markdown headings (#, ##, ###) from body. Use semantic XML tags.

Required tags (every skill must have):

<objective> - What the skill does and why it matters
<quick_start> - Immediate, actionable guidance
<success_criteria> or <when_successful> - How to know it worked

Conditional tags (add based on complexity):

<context> - Background information
<workflow> or <process> - Step-by-step procedures
<advanced_features> - Deep-dive topics
<validation> - How to verify outputs
<examples> - Multi-shot learning
<anti_patterns> - Common mistakes
<security_checklist> - Security patterns
<testing> - Testing workflows
<common_patterns> - Code examples
<reference_guides> - Links to reference files

Tag selection intelligence:

Simple skills: Required tags only
Medium skills: Required + workflow/examples
Complex skills: Required + conditional tags as needed

Writing style:

Use imperative/infinitive form (verb-first instructions)
Not second person ("you should")
Objective, instructional language
Example: "To accomplish X, do Y" not "You should do X" </step_4>

<step_5> Adding Bundled Resources

<scripts_directory> When to use scripts:

Same code rewritten repeatedly
Deterministic reliability needed
Token efficiency important

Benefits:

More reliable than generated code
Save tokens (no code in context)
Ensure consistency across uses

Organization:

skill-name/
├── SKILL.md
├── scripts/
│   ├── main_utility.py
│   ├── helper_script.py
│   └── validator.py
└── references/

Script documentation pattern:

<utility_scripts>
**analyze_form.py**: Extract all form fields from PDF

```bash
python scripts/analyze_form.py input.pdf > fields.json

Output format:

{
  "field_name": { "type": "text", "x": 100, "y": 200 }
}

validate_boxes.py: Check for overlapping bounding boxes

python scripts/validate_boxes.py fields.json
# Returns: "OK" or lists conflicts

</utility_scripts>


**Script best practices**:
- Handle errors gracefully (don't punt to Claude)
- Document configuration values (avoid "voodoo constants")
- List required packages in SKILL.md
- Make clear if Claude should execute or read as reference
</scripts_directory>

<references_directory>
**When to use references**:
- Detailed patterns and techniques
- Comprehensive API documentation
- Migration guides
- Edge cases and troubleshooting
- Extensive examples

**Benefits**:
- Keeps SKILL.md lean
- Loaded only when Claude determines it's needed
- Can be large (2,000-5,000+ words each)

**Organization**:
- Keep references one level deep from SKILL.md
- Add table of contents for files >100 lines
- Use pure XML structure in reference files

**Reference pattern in SKILL.md**:
```xml
<reference_guides>
For detailed patterns and techniques:

- **`references/patterns.md`** - Common patterns and best practices
- **`references/api-reference.md`** - Complete API documentation
- **`references/advanced.md`** - Advanced use cases and edge cases
</reference_guides>

</references_directory>

<assets_directory> When to use assets:

Templates, images, icons
Boilerplate code
Sample documents

Benefits:

Separates output resources from documentation
Enables Claude to use files without loading into context

Examples:

assets/logo.png for brand assets
assets/slides.pptx for PowerPoint templates
assets/frontend-template/ for HTML/React boilerplate </assets_directory> </step_5>

<step_6> Validation and Testing

Structure validation:

YAML frontmatter valid (name matches directory)
Description in third person with specific triggers
Pure XML body (no markdown headings)
Required tags present (objective, quick_start, success_criteria)
All XML tags properly closed
SKILL.md under 500 lines
Referenced files exist

Content validation:

Imperative/infinitive form (not second person)
Clear, specific instructions
Examples included when format matters
Success criteria defined
Progressive disclosure applied

Testing:

Test with real usage scenarios
Verify skill triggers on expected user queries
Check that references load when needed
Iterate based on observations </step_6>

<step_7> Creating Slash Command Wrapper (Optional)

Create a lightweight wrapper for discoverability:

Location: ~/.claude/commands/{skill-name}.md

Template:

---
description: {Brief description}
argument-hint: [{argument description}]
allowed-tools: Skill({skill-name})
---

<objective>
Delegate {task} to the {skill-name} skill for: #$ARGUMENTS

This routes to specialized skill containing patterns, best practices, and workflows.
</objective>

<process>
1. Use Skill tool to invoke {skill-name} skill
2. Pass user's request: #$ARGUMENTS
3. Let skill handle workflow
</process>

<success_criteria>
- Skill successfully invoked
- Arguments passed correctly to skill
</success_criteria>

The slash command's only job is routing—all expertise lives in the skill. </step_7>

<step_8> Iteration and Improvement

After testing, iterate based on real usage:

Use the skill on real tasks
Notice struggles or inefficiencies
Identify improvements needed
Implement changes and test again

Common improvements:

Strengthen trigger phrases in description
Move long sections to references/
Add missing examples or scripts
Clarify ambiguous instructions
Add edge case handling </step_8> </skill_creation_process>

<naming_conventions> <verb_noun_pattern> Use verb-noun convention for skill names:

create-* - Building/authoring tools

Examples: create-agent-skills, create-hooks, create-landing-pages

manage-* - Managing external services

Examples: manage-facebook-ads, manage-zoom, manage-stripe

setup-* - Configuration/integration

Examples: setup-stripe-payments, setup-meta-tracking

generate-* - Generation tasks

Examples: generate-ai-images, generate-reports

Avoid:

Vague: helper, utils, tools
Generic: documents, data, files
Reserved: anthropic-helper, claude-tools </verb_noun_pattern>

<validation_rules>

Maximum 64 characters
Lowercase letters, numbers, hyphens only
No XML tags
No reserved words: "anthropic", "claude"
Must match directory name exactly </validation_rules> </naming_conventions>

<anti_patterns> ❌ Bad: Using markdown headings in body

# PDF Processing
## Quick start

✅ Good: Pure XML structure

<objective>PDF processing...</objective>
<quick_start>...</quick_start>

✅ Good: Specific trigger phrases

description: This skill should be used when the user asks to "create a hook", "add a PreToolUse hook", or mentions hook events.

✅ Good: Progressive disclosure

skill-name/
├── SKILL.md  (1,500 words)
└── references/
    ├── patterns.md (2,500 words)
    └── advanced.md (3,700 words)

✅ Good: Imperative form

Start by reading the configuration file.
Validate the input before processing.

✅ Good: Clear resource references

<reference_guides>
- **`references/patterns.md`** - Detailed patterns
- **`references/advanced.md`** - Advanced techniques
</reference_guides>

<best_practices_summary> DO:

Use third-person in description ("This skill should be used when...")
Include specific trigger phrases ("create X", "configure Y")
Keep SKILL.md lean (under 500 lines)
Use progressive disclosure (move details to references/)
Write in imperative/infinitive form
Apply create-prompt prompting techniques
Reference supporting files clearly
Provide working examples
Create utility scripts for common operations
Test with real usage

DON'T:

Use second person anywhere
Have vague trigger conditions
Put everything in SKILL.md (>500 lines without references/)
Use markdown headings in body
Leave resources unreferenced
Include broken or incomplete examples
Skip validation
Forget to apply prompting best practices </best_practices_summary>

<reference_guides> For deeper topics, see reference files:

references/script-patterns.md - Executable code patterns, error handling, package dependencies
references/progressive-disclosure-patterns.md - Advanced progressive disclosure strategies
references/xml-tag-guide.md - Complete guide to XML tags and structure
references/prompting-integration.md - How to integrate create-prompt techniques into skills
references/real-world-examples.md - Complete examples of well-structured skills </reference_guides>

<success_criteria> A well-structured skill has:

Valid YAML frontmatter with third-person description and specific triggers
Pure XML structure with no markdown headings in body
Required tags: objective, quick_start, success_criteria
Conditional tags appropriate to complexity level
Progressive disclosure (SKILL.md < 500 lines, details in references/)
Clear, concise instructions in imperative form
Expert prompting techniques from create-prompt applied
Bundled resources (scripts, references, assets) as needed
Clear references to all supporting files
Real-world testing and iteration
Optional slash command wrapper for discoverability </success_criteria>

create-skills

SKILL.md

スコア

レビュー